Research map / Large language models / Language, safety and interpretability
CLadder: Assessing Causal Reasoning in Language Models
Zhijing Jin, Yuen Chen, Felix Leeb, Luigi Gresele, Ojasv Kamal, Zhiheng LYU, Kevin Blin, Fernando Gonzalez Adauto, Max Kleiman-Weiner, Mrinmaya Sachan, Bernhard Schölkopf
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Apply six-step causal-inference chain-of-thought prompting instead of direct prompting | GPT-4 · CLadder v1.5 | accuracy: 62.03 -> 70.40; 8.37 points | Improved |
| Apply six-step causal-inference chain-of-thought prompting instead of direct prompting | GPT-4 · CLadder v1.5, rung 1 | accuracy: 63.01 -> 83.35 | Improved |
| Apply six-step causal-inference chain-of-thought prompting instead of direct prompting | GPT-4 · CLadder v1.5, rung 2 | accuracy: 62.82 -> 67.47 | Improved |
| Apply six-step causal-inference chain-of-thought prompting instead of direct prompting | GPT-4 · CLadder v1.5, rung 3 | accuracy: 60.55 -> 62.05 | Improved |
| Apply six-step causal-inference chain-of-thought prompting instead of direct prompting | GPT-4 · CLadder v1.0 | accuracy: 64.28 -> 66.64; 2.36 points | Improved |
| Use CLadder CoT on anti-commonsensical examples rather than vanilla GPT-4 | GPT-4 · CLadder v1.5, anti-commonsensical | accuracy: 60.47 -> 70.12; 9.65 points | Improved |
| Remove Step 1, causal graph extraction (ablation) | CLadder CoT · CLadder v1.0 | accuracy: 66.64 -> 64.54 | Hurt |
| Remove Step 2, query-type identification (ablation) | CLadder CoT · CLadder v1.0 | accuracy: 66.64 -> 63.74 | Hurt |
| Remove Step 3, query formalization (ablation) | CLadder CoT · CLadder v1.0 | accuracy: 66.64 -> 63.43 | Hurt |
| Remove Step 4, available-data extraction (ablation) | CLadder CoT · CLadder v1.0 | accuracy: 66.64 -> 64.47 | Hurt |
Closest papers by meaning
- Unveiling Causal Reasoning in Large Language Models: Reality or Mirage? (NeurIPS 2024 · similarity 0.93)
- Can Large Language Models Infer Causation from Correlation? (ICLR 2024 · similarity 0.93)
- COLD: Causal reasOning in cLosed Daily activities (NeurIPS 2024 · similarity 0.93)
- Mitigating Hallucinations in Large Language Models via Causal Reasoning (AAAI 2026 · similarity 0.90)
- Nuance Matters: Probing Epistemic Consistency in Causal Reasoning (AAAI 2025 · similarity 0.90)
- Causality can systematically address the monsters under the bench(marks) (NeurIPS 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
